Grok’s Imagine tool is an AI-powered image creator that can turn text prompts into pictures, transform existing images into short videos, and handle a variety of advanced image-editing tasks. Unfortunately, this tool has serious safety and ethical concerns because it allows users to generate sexual and semi-sexual imagery through its dedicated “Spicy Mode.”

A now-deleted post from xAI employee Mati Roy said, “Grok Imagine videos have a spicy mode that can do nudity,” and said in another post on X that it would “be able to create realistic videos of humans.”

Independent reporting confirms that the feature can produce erotic or semi-nude images and even short animated sexual videos. In hands-on testing, journalists were able to generate sexualized content with minimal friction, noting that while some explicit details may be blurred, the system still produces imagery clearly designed for erotic consumption.

Social media users, on the other hand have posted sexual short videos that depict female nudity (breasts and/or buttocks) and male genitalia, with motions that suggest intercourse. While someone could technically argue that the output falls just short of outright pornography, it’s just as reasonable to say that it is pornography in everything but name. And in practice, that distinction doesn’t matter—especially when there is serious risk that identifiable individuals can be attached to or implicated in this kind of content.

In short, Grok is playing semantics with privacy and user safety.

And the tool’s moderation guardrails appear inconsistent, with some sexually explicit prompts passing through while milder or artistic material is blocked, leaving users and potential victims exposed to unpredictable—and unsafe—outcomes.

NOTE: Censored proof of Grok Imagine videos depicting nude AI characters evoking penetration are available to journalists or policymakers upon request.

Consumer Federation of America (CFA) was joined by NCOSE and other privacy and child safety organizations in sending a joint letter to Attorneys’ General of the United States, United States Attorneys’ Offices, and Federal Trade Commission regarding Grok Imagine. The letter noted: “As of testing on August 11th, the platform does not offer the ‘spicy option’ for real photos uploaded by users, but still generates nude videos from images generated by the tool, which can be used to create images that look like real, specific people.”

Unfortunately, this is not theoretical. It’s a documented reality. Investigation by The Verge discovered:

“it didn’t hesitate to spit out fully uncensored topless videos of Taylor Swift the very first time I used it — without me even specifically asking the bot to take her clothes off.” 

The BBC noted that it had seen “several examples on the social media platform X of people asking the chatbot to undress women to make them appear in bikinis without their consent, as well as putting them in sexual situations.” The BBC shared the story of Samantha, who described feeling “dehumanised” after Grok was used to digitally remove her clothing. When Samantha posted about her experience on X, several other users commented about having experienced the same abuse.

One blogger who was victimized through deepfake pornography made with Grok chillingly described the traumatic impact:

“I have never felt more like a paper doll than I have felt in the last 72 hours. I have felt dirty, like a ghost, avoiding eye contact with friends in the street, convinced that everyone who sees me now sees that image.”

She expressed horror that “this is the world we live in now, a world where this is something men can do easily and en masse, to women existing peacefully on the internet.” And unfairness that women are experiencing “[a]ll this, because of someone who will never have to meet our eyes, who probably does not even know our names.”

Grok has made this kind of anonymized, detached, mass-scale violations of women easy.

Even worse, the nonconsensual images Grok generates are populating into the media gallery on Grok’s public X profile. In other words: Grok is not only generating them for the user, but posting them for all the world to see. International Business Times called it a “public archive of nudes.” 

There are also ethical problems with how Grok trains its image generator. The Consumer Federation of America letter noted: 

“Furthermore, image generation platforms train off of scraped and licensed publicly available data including untold amounts of photos of real people. According to one study, one popular image training dataset contained 102 million images of real people from photos on school sites, LinkedIn, Flickr, and more, even after attempts at “data sanitization.” xAI does not disclose the contents of its training dataset, but that dataset likely contains a large number of real photos. xAI has incentivized users to upload sensitive data for one purpose, and then use it to train or for other purposes, and this practice likely applies to photos as well. When photos of people are included in a dataset used for AI training, it increases the likelihood of an image representation of that person being spat out by the AI generator. This, in turn, makes it more likely that a photo of you posted on X will be hoovered up by Grok’s system and integrated into its training dataset.

When a mainstream, Teen-rated platform enables the generation of sexual content with few barriers, it creates fertile ground for abuse, harassment, exploitation, and reputational harm. Grok Imagine’s current configuration fails to meet even baseline expectations for responsible AI governance.